Machine Learning in Bioinformatics

CoMR: an integrative scoring pipeline for Comprehensive Mitochondrial proteome Reconstruction across eukaryotes

CoMR improves mitochondrial proteome reconstruction by integrating multiple evidence sources, achieving better performance than standalone predictors in both model and non-model eukaryotes.

Can Large Language Models ''Read'' Biological Sequences? A Systematic Evaluation of In-Context Learning for Antibody Characterization

This study evaluates how large language models perform in-class learning for antibody classification using biological sequences, finding that few-shot prompting with diverse demonstrations achieves competitive results.

Preferential CDR masking in paired antibody language models improves binding affinity prediction

This paper shows that focusing on complementarity-determining regions (CDRs) in antibody sequences improves binding affinity prediction using region-aware pretraining strategies.

Decoding protein language models: insights from embedding space analysis

This paper introduces a framework for comparing embedding spaces in molecular biology, demonstrating its utility in pre-filtering machine learning tasks and identifying differential patterns across embeddings.

Learning Shared Residue Backgrounds and Modification-Specific Offsets for PTM Site Prediction

The paper introduces a framework for multi-type PTM site prediction using protein language model embeddings, achieving improved performance in a shared-model setting.

A Practice on Antibody Hydrophobic Interaction Chromatography Retention Time Prediction using Pre-Trained Large Language Model Fine-Tuning

This study fine-tunes a pre-trained language model to predict antibody hydrophobicity using retention time data, achieving high accuracy with rigorous data quality control.

GFMBench-API: A Standardized Interface for Benchmarking Genomic Foundation Models

Wrap a genomic foundation model's inference methods once to score it on supervised and zero-shot GUE, VariantBenchmarks, BEND, TraitGym and ClinVar tasks; code is promised upon acceptance.

Data-Centric Evaluation of Protein Function Prediction Pipelines

The paper evaluates how data preprocessing steps affect protein function prediction performance using antioxidant protein classification.

ProtJEPA: A Multimodal Joint-Embedding Predictive Architecture for Protein Biological World Modeling with Multi-TeacherModality-Attentive Fusion

ProtJEPA predicts joint embeddings across ten biological modalities from protein sequences, outperforming baselines in zero-shot GO retrieval and enzyme class prediction.

Evaluating Lightweight and Full Fine-Tuning Strategies Against Classical Machine Learning for Protein Function Prediction

This study compares different fine-tuning strategies and classical machine learning approaches for protein function prediction using benchmark datasets.

Mapping the Competence Boundary of a Protein Property Model: A Case Study on Plastic-Degrading Enzymes using ProtTrust-XAI

The paper introduces ProtTrust-XAI, a framework to assess model trustworthiness for protein properties, showing it can identify reliable predictions and detect distribution shifts in plastic-degrading enzymes.

The N-Terminus of Sophora tonkinensis Cytochrome P450s Evolves Neutrally yet Encodes Rich Functional Information: A Protein Language Model Analysis

The paper uses a protein language model to analyze the N-terminal region of Sophora tonkinensis cytochrome P450s, revealing neutral evolution and functional encoding through two separable signals.

PLMView: collaborative protein language model representations for fast and scalable specialized protein function inference

PLMView is an unsupervised framework for fine-grained protein function classification from sequence, achieving high precision and speed without labeled data or model fine-tuning.

Graph Machine Learning for Physiological Role Prediction in Protein Contact Networks: A Large-Scale Comparative Study on the Human Proteome

This paper compares graph machine learning methods for predicting protein function from contact networks, finding GNNs outperform kernel methods in EC class prediction.

Interpreting Protein Language Models: high attention sites predict functional regions

The paper identifies high attention sites in protein language models that predict functional and pathogenic regions in proteins.

CPPLocPred: Subcellular Localization of Cell-Penetrating Peptides

CPPLocPred predicts subcellular localization of cell-penetrating peptides using a hierarchical machine learning framework, achieving high accuracy on validation datasets.

A curated human lactylome and protein language model framework enable accurate prediction and reveal local determinants of lysine lactylation

The paper introduces CLEAR-Lactyl, a curated dataset of human lysine lactylation sites, and AttentionKla, a deep learning model that accurately predicts lactylation sites and reveals sequence determinants.

Cross-Attention Over RNA And Protein Sequences Enables Generalizable Interaction Prediction

The paper introduces CORAL, a deep learning framework for RNA-protein interaction prediction that improves generalization and interpretability through cross-attention mechanisms.

pLM representations unlock metagenomic space beyond homology

This paper shows that protein language models can better annotate metagenomic proteins when trained on curated data, enabling retrieval of diverse enzyme candidates from billions of sequences.

ViralMap: Predicting Features in Viral Proteins from Primary Sequence

ViralMap predicts ten annotation classes for viral proteins from primary sequences, achieving high performance on seven classes and generalizing across viral strains and families.

ProtSyntax: a protein large language model for decoding post-translational modification syntax and function

ProtSyntax is a protein language model that improves PTM site prediction and functional inference by integrating contextual and structural information.

X-PAIR: an ultrafast multitask framework for proteome-scale reconstruction of PPI networks and partner-specific interfaces from sequence

X-PAIR is a multitask deep learning framework that predicts protein-protein interactions and identifies interface residues from sequence data, outperforming existing methods and enabling proteome-scale analysis.

High-resolution dissection of concept acquisition in different families of protein language models

This study analyzes how biological concepts are encoded in protein language models across their layers and finds that complexity increases with depth, with pretraining data and compute having a greater impact than parameter scaling.

BioReason-Pro: Advancing Protein Function Prediction with Multimodal Biological Reasoning

A reasoning LLM combining ESM3 embeddings with GO-GPT term predictions produces protein function annotations that experts rated tie-or-better than UniProt in 79% of cases (SFT model).

Systematic contextual biases in SegmentNT potentially relevant to other nucleotide transformer models

The paper identifies contextual biases in the SegmentNT model, showing how input sequence length and nucleotide position affect predictions and suggests ways to mitigate these biases.

A machine learning model predicts protein stability of annotated and alternate protein isoforms

This study develops a machine learning model to predict protein stability from N-terminal sequences and identifies stability-determining motifs in annotated and alternative isoforms.

MINA: linear probes reveal coding-sequence family signal in frozen DNA encoders up to protein composition

The paper evaluates frozen DNA encoders' ability to recover protein-family labels and GenePT embeddings from coding sequences and TSS contexts, finding limited success in gene-function prediction.

TEDlm: domain-centric protein language models with optional structural pre-training

TEDlm is a domain-centric protein language model that improves remote-homology detection and Molecular Function prediction by pretraining on structurally-defined domain sequences.

Asymmetric Structural Transfer Between Natural Language and Biological Foundation Models

The paper investigates whether structural transfer between natural language and biological foundation models is directional, finding that transfer from language to biology is stronger than the reverse.

BioMetAll v2.0: Introducing Scores, Metal Discrimination, and Side-Chain Descriptors for Predicting Metal-Binding Sites in Proteins.

BioMetAll v2.0 improves metal-binding site prediction with metal-specific scores and backbone geometry analysis, enhancing benchmarking and understanding of metal-protein interactions.